NanoClaw

NanoClaw is a lightweight open-source personal AI agent that focuses on understandability, isolation, and user-owned customization. Its public site frames the project as a smaller alternative to OpenClaw: a small source tree, a single Node host process, per-agent containers, channel adapters installed as skills, and credential handling through OneCLI’s Agent Vault. It is not trying to be a giant framework; it is trying to be a personal assistant stack one person can inspect and modify.

Top features

• Small codebase: the project emphasizes a compact implementation that a user can realistically read and customize.
• Container isolation: each agent group runs in its own Docker container, with optional Docker Sandboxes micro-VM isolation or Apple Container on macOS.
• Multi-channel messaging: WhatsApp, Telegram, Discord, Slack, Microsoft Teams, iMessage, Matrix, Google Chat, Webex, Linear, GitHub, WeChat, and email are listed as supported channels.
• Flexible routing: channels can map to separate agents for privacy, share one agent for unified memory, or fold multiple channels into one session.
• Credential security: agents do not hold raw API keys; outbound requests route through OneCLI’s Agent Vault for injection and policy control.
• Skill-installed extensions: channels and provider adapters are added on demand rather than shipping every integration in trunk.

Use cases

• Run a self-owned personal assistant from messaging apps while keeping each agent in an isolated container.
• Build a small OpenClaw-style setup that can be audited without weeks of framework study.
• Keep work, personal, and experimental assistants separated by agent group, workspace, memory, and mounts.
• Customize the assistant through Claude Code rather than learning a large plugin framework first.
• Compare secure-by-isolation personal agents against OpenClaw, Hermes, PicoClaw, and ZeroClaw.